Cairn is an open-source (AGPLv3) coordination layer for multi-agent AI systems. Agents coordinate through a passive, append-only shared blackboard and minimal per-agent context instead of conversation, so they stop silently overwriting each other's shared state, and they do it with roughly half the tokens (a measured ~54% fewer tokens / ~65% lower cost on an identical task). On Celo, Cairn adds verifiability: every coordination run can emit an on-chain receipt and each agent carries an ERC-8004 on-chain identity, turning ephemeral multi-agent work into auditable, trustless coordination. It is reusable infrastructure other agent projects build on (chain-agnostic Identity / Receipt / Payment adapters), live on Celo mainnet.
Make multi-agent AI coordination cheap, debuggable, and verifiable by default — give agents a shared, append-only memory instead of conversation, and an on-chain receipt + identity layer so independent agents can coordinate and transact with auditable trust. Open infrastructure (AGPLv3) for the emerging agent economy.
Multi-agent AI systems have no shared-state coordination layer. When several agents work together, they coordinate through conversation and silently overwrite each other's shared state — a split-brain, undebuggable failure mode that teams keep hand-rolling a custom event log to patch. The dominant agent protocols (A2A, MCP) explicitly leave this shared-state layer out. Two costs follow: (1) coordination burns large token budgets (conversation-heavy context passed every turn), making multi-agent systems expensive and brittle; and (2) when agents owned by different parties must coordinate or transact, there is no verifiable, auditable record of what each agent did — no trust primitive for an open agent economy.
Cairn replaces agent conversation with a passive, append-only shared blackboard + minimal per-agent context: every contribution is a durable, auditable trace, nothing is overwritten, and coordination costs roughly half the tokens (measured ~54% fewer tokens / ~65% lower cost on an identical task). On Celo it adds verifiability — each run can emit an on-chain coordination receipt and each agent carries an ERC-8004 on-chain identity — so multi-agent work becomes auditable and trustless. AGPLv3, reusable infrastructure other agent projects build on.
